Attribution, Not Authority began as my independent study of when reasoning models adopt false suggestions. I built a Python evaluation harness, analyzed the reasoning traces, and reported the conditions that changed adoption—including an unresolved discrepancy in a replication test.
For my first-author IEEE paper, I measured a deployed vision-language pipeline stage by stage across 180 events, finding that configured delay and remote inference—not transport—accounted for most of the detection cycle. I also built a programming language, compiler, and virtual machine from scratch, and published SmartCompress FS on Ubuntu’s Snap Store.
